%rb = tfrt_fallback_async.const_dense_tensor dense<0.5> : tensor<f32> {_tfrt_cost = 1 : i64}

  // CHECK: tfrt_fallback_async.set_resource {{%.*}}, {{%.*}} {device = "cpu", index = 0 : i64}
  // CHECK: tfrt_fallback_async.set_resource {{%.*}}, {{%.*}} {device = "cpu", index = 1 : i64}
  // CHECK: tfrt_fallback_async.get_resource {{%.*}} {_tfrt_cost = 1 : i64, device = "cpu", indices = [0, 1]}

  %2 = stablehlo.convolution(%arg0, %0) dim_numbers = [b, f, 0, 1]x[o, i, 0, 1]->[b, f, 0, 1], window = {pad = [[1, 1], [1, 1]]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<1x8x4x4xf32>, tensor<8x8x3x3xf32>) -> tensor<1x8x4x4xf32>
  return %2 : tensor<1x8x4x4xf32>
}

// CHECK-DAG: %[[CONST:.+]] = stablehlo.constant {{.*}} : tensor<8x8x3x3xf32>

      lhs_quantization_axis = -1 : i64,
      lhs_quantization_min_val = -2147483648 : i64,
      lhs_quantization_max_val = 2147483647 : i64,
      rhs_quantization_axis = -1 : i64,
      rhs_quantization_min_val = -2147483648 : i64,
      rhs_quantization_max_val = 2147483647 : i64,
      output_quantization_axis = -1 : i64,
      output_quantization_min_val = -2147483648 : i64,

    "tf.InitializeTableFromTextFileV2"(%0, %arg0) {delimiter = "\09", device = "", key_index = -2 : i64, offset = 0 : i64, value_index = -1 : i64, vocab_size = 437 : i64} : (tensor<!tf_type.resource>, tensor<!tf_type.string>) -> ()

{_xla_inferred_shapes = [#tf_type.shape<*>, #tf_type.shape<*>], allowed_batch_sizes = [64, 128, 256], batch_timeout_micros = 5000 : i64, batching_queue = "", container = "", device = "/job:localhost/replica:0/task:0/device:CPU:0", enable_large_batch_splitting = true, f = @compute_0, max_batch_size = 256 : i64, max_enqueued_batches = 10000 : i64, num_batch_threads = 2 : i64, operandSegmentSizes = array<i32: 1, 1>, shared_name = "computation"} : (tensor<?x?xi32>, tensor<?x?xi32>) -> (tensor<*xi32>, tensor<*xi32>)...

  %57 = "tf.Const"() {value = dense<0> : tensor<4xi32>} : () -> tensor<4xi32>
  %534 = "tf.StridedSlice"(%arg0, %57, %57, %18) {begin_mask = 11 : i64, device = "", ellipsis_mask = 0 : i64, end_mask = 11 : i64, new_axis_mask = 4 : i64, shrink_axis_mask = 0 : i64} : (tensor<*xf32>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>) -> tensor<*xf32>
  "func.return"(%534) : (tensor<*xf32>) -> ()

  "tf.InitializeTableFromTextFileV2"(%0, %cst) {delimiter = " ", device = "", key_index = -1 : i64, value_index = -1 : i64, vocab_size = -1 : i64} : (tensor<!tf_type.resource>, tensor<!tf_type.string>) -> ()
  func.return
  // CHECK: [[VAL:%.*]] = "tf.HashTableV2"()

    %VarHandleOp, %ctl = VarHandleOp device("/CPU:0") name("x") {_output_shapes = [#tf_type.shape<>], allowed_devices = [], container = "a", dtype = i64, shape = #tf_type.shape<>, shared_name = "x"} : () -> (tensor<!tf_type.resource<tensor<i64>>>)

  // CHECK: _arithmetic_count = 1000 : i64
  %0 = "tfl.add"(%arg0, %arg1) {fused_activation_function = "NONE"} : (tensor<10x10x10xf32>, tensor<10x10x10xf32>) -> tensor<10x10x10xf32>
  func.return %0 : tensor<10x10x10xf32>
}

func.func @testAddBroadcast(%arg0: tensor<10x10x10xf32>, %arg1: tensor<10xf32>) -> tensor<10x10x10xf32> {
  // CHECK: _arithmetic_count = 1000 : i64

module attributes {tf_saved_model.semantics} {
  func.func @no_call_once(%arg0: tensor<i64> {tf_saved_model.index_path = ["x"]}) -> (tensor<i64> {tf_saved_model.index_path = ["r"]})
  attributes {tf.entry_function = {control_outputs = "", inputs = "input:0", outputs = "output:0"}, tf_saved_model.exported_names = ["serving_default"]} {
    func.return %arg0 : tensor<i64>
    // CHECK-LABEL: no_call_once
    // CHECK-NOT: "tfl.call_once"
  }